iATPSnFR2: A high-dynamic-range fluorescent sensor for monitoring intracellular ATP

Significance Adenosine triphosphate (ATP) is a key metabolite necessary for cellular life. Here, we develop a next-generation genetically encoded ratiometric fluorescent ATP sensor that allows subcellular tracking of ATP levels in living cells. The large dynamic range makes it possible to follow the dynamics of this metabolite across cells and subcellular regions under different metabolic stressors. We expect that iATPSnFR2, combined with proper controls for assessing changes in pH, will provide researchers with exciting opportunities to study ATP dynamics with high temporal and spatial resolution.

Adenosine triphosphate (ATP) is a critical biochemical currency.Its hydrolysis to adenosine diphosphate (ADP) provides the free energy necessary to drive numerous physiological processes.The importance of ATP is evident by the plethora of routes that exist to convert the energy stored in the form of combustible hydrocarbons into ATP.The molecular details of glycolysis and oxidative phosphorylation have been established for over 60 y, having been successfully dissected through in vitro enzymology and biochemistry.One trade-off with this type of reductionism is that it is often hard to recapitulate the physiological milieu in vitro, and therefore, the nuances of how molecular pathways are regulated in real time in specific subcellular locations are missed.Fluorescent biosensors are designed to fill the knowledge gap between in vitro reconstitution biochemistry and cellular physiology as they can reveal subcellular dynamics of different metabolites or ions with high temporal and spatial resolution in living cells and tissues.Successful deployment of a fluorescent biosensor requires that the affinity of the sensor matches the physiological concentration of its cognate analyte and that the sensor discriminates against chemically related species.The signal-to-noise ratio (SNR) is determined by the ratio of the sensor's fluorescence in the bound and unbound states.SNR defines the magnitude of change in analyte concentration that can be detected distinctly from other sources of signal fluctuation.Higher SNR reduces the spatial scale over which the signal must be averaged.Several promising strategies have emerged in the last 10 y to develop a genetically encoded sensor for ATP.
An intensity-based ATP Sensing Fluorescent Reporter, iATPSnFR (1), was developed by inserting circularly permuted superfolder GFP between the two ATP-binding helices of the epsilon subunit of the F 0 -F 1 ATPase of thermostable Bacillus subtilis PS3, which undergoes a large conformational change upon binding ATP (2).Other research groups have made similarly conceived sensors, albeit with different topologies, ATP affinities, and dynamic ranges.These include ChemoG-ATP SiR (3), MaLion (4), and QUEEN (5), which, like iATPSnFR, have a common ancestor in "ATeam" (6).ATeam is a FRET-based sensor in which cyan and yellow fluorescent proteins [or GO-ATeam, using GFP/orange fluorescent protein (7)] were placed at the N and C termini of the bacterial ε subunit of F 0 -F 1 ATPase; their FRET efficiency increases through ATP-induced conformational change.Most of these cytosolic ATP sensors have been reviewed recently (8).
An advantage of sensors based on soluble ligand binding domains is that they can be targeted to subcellular locations of interest.ATeam has been targeted to the mitochondria and used to observe histamine-induced changes in [ATP] in HeLa cells (7) and to distinguish the relative contributions of glycolysis/oxidative phosphorylation to ATP production in different cell lines at the cellular level (9).It has also been used to study the dynamics of ATP consumption at synapses (see Discussion).

pnas.org
Unfortunately, each of these sensors has been lacking in some respect.QUEEN detects changes in ATP by reporting different excitation wavelengths, which is not practical or useful for most imaging experiments, especially for preparations other than monolayers.MaLion is available as either a green or red fluorescent sensor and has almost fivefold increase in fluorescence upon saturation with ATP but loses dynamic range at higher temperatures.The most recently developed ATP sensor, ChemoG-ATP SiR , is an improved chemogenetic FRET sensor with 10-fold change in fluorescence ratio and a physiologically appropriate affinity of about 2 mM (3) .It also is available in different colors, depending on the FRET acceptor used.However, the use of two fluorophores by Förster resonance energy transfer (FRET)-based sensors occupies spectral bandwidth that might constrain its application to multiplex imaging, and requires more complicated optics when imaging at higher frame rates.ChemoG-ATP SiR is further hindered by the limited availability of the custom-made HaloTagLigand-SiR fluorophore.It will also be prone to misreporting ATP concentrations across cells if saturation of the sensor protein with the SiR-HaloTag ligand is incomplete.I.e., different red-FRET:green ratios could be interpreted as different concentrations of ATP, when in reality they might be the result of unequal loading of the synthetic fluorophore.Control experiments are necessary to ensure that full saturation of the HaloTag is complete before reliable FRET measurements can be made.Other ATP sensors based on the bacterial ATP regulatory domain GlnK1, including Perceval (10) and PercevalHR (11), have similar affinities for ATP and ADP, and thus provide a measure of the ratio of ATP:ADP more than the concentration of ATP alone.Finally, firefly luciferase can be used to measure the concentration of ATP, as there is a direct relationship between ATP consumption and light production.This approach is obviously limited in its spatial and temporal resolution but was successfully adapted to observe ATP consumption at nerve terminals in primary dissociated neurons (12).Even so, that sensor is difficult to deploy due to the low photon flux and limited applicability to modern optical sectioning.
Here, we report the improvement of iATPSnFR1 (1).iATPSnFR2 has a much higher dynamic range (ΔF/F ~ 12), is available as three different affinity variants, and is fused to spectrally separable fluorescent tags (HaloTag with synthetic far-red fluorophores or mIR-FP670nano3), thus providing an approach to normalize the signal to the expression level of the sensor and allowing quantitative comparisons across individual cells or subcellular locations.We show that iATPSnFR2 can provide detailed measurements of the variations in resting ATP values across synapses as well as the kinetics of ATP changes during metabolic perturbations at the cytosolic, single-synapse, and single-mitochondrion levels.These data show that individual synapses behave as semi-independent metabolic units, as during metabolic stress, the kinetics of ATP depletion varied significantly even within the same axon.

Results
Sensor Design and In Vitro Characterization.To improve upon iATPSnFR1, we reevaluated the composition of the linkers connecting the ATP-binding domain and the cpSFGFP (Fig. 1A).This was done with the inclusion of an ATP-affinity boosting mutation, A95K, as we were also aiming to create a high-affinity ATP sensor that would be useful for the detection of submicromolar amounts of extracellular ATP.To increase ΔF/F we screened thousands of variants of the sensor in bacterial lysate, as described previously (1), but expanded the regions mutated to cover a larger number of residues.The sensor with the highest maximum ΔF/F (saturated vs. unbound) has the residues comprising "linker 1" as VLVG, where the residues QD adjacent to the ATP-binding domain are changed to VL, and the residues SH adjacent to cpSFGFP are changed to VG.Additional residues ICV were placed in "linker 2" between the end of cpSFGFP and residue 110 of the ATP-binding domain.This sensor (L1-VLVG, L2-ICV, A95K) has an affinity for ATP of ~16 µM.
To tune the affinity of the sensor into the range needed for measuring cytosolic ATP in neurons [estimated to be ~1.4 mM ( 12)], we reverted A95K back to alanine, which also reduced ΔF/F.We then screened residues near the ATP-binding site for variants that had both a high ΔF/F upon saturation with ATP and a K d near 1.4 mM.The substitution A119L satisfied those two criteria and was carried forward as iATPSnFR2 (L1-VLVG, L2-ICV, A95A, A119L).In parallel, we screened other amino acid positions in the epsilon subunit in the context of A95K for variants that bound ATP more tightly.We found that the double mutant S29W.A95K has an affinity for ATP of ~4 µM, giving us a trio of sensors with similar ΔF/F and varying affinities: S29W.A95K (4 µM), A95K (16 µM), and A95A.A119L (530 µM).(Fig. 1B and SI Appendix, Fig. S1A).
Like GCaMP and other cpGFP-based sensors, the excitation peaks at 405 nm and 485 nm are shifted relative to each other upon saturation with ligand, providing an isosbestic point at 436 nm (SI Appendix, Fig. S1 B and C), which provides an ATP-independent signal that may be useful for normalization of focus and movement artifacts.However, since imaging with shorter wavelength excitation light is often detrimental to live cells, we also developed versions of the sensor that include a C-terminal fusion protein to act as a normalization reference.A particularly useful C-terminal fusion protein is HaloTag, which can serve as a conjugation partner for synthetic fluorophores (13).There are many synthetic fluorophores available for conjugation to HaloTag; for the present study, we primarily used the far-red dyes JF635-or JFX650-HaloTag ligand (14).
The iATPSnFR2.HaloTag-JFX650 conjugate was the primary protein used for in vitro characterization of performance and specificity.The inclusion of HaloTag affects affinity and also appears to increase the shelf-life of the purified sensor.iATPSnFR2.A95A.A119L.HaloTag-JFX650 has a ΔF/F of ~12 and K d for ATP of 500 µM (Fig. 1B), when taken at the maximum point of excitation/emission sensitivity.It has low affinity (5.3 mM) and ΔF/F (~6) for ADP (Fig. 1B and SI Appendix, Fig. S1A).This version of the sensor has minimal, if any, affinity for AMP.It also changes fluorescence in response to GTP and CTP, but not TTP (SI Appendix, Fig. S1D).The affinity for these nucleoside triphosphates is low enough that their presence in cells should not affect sensor performance, given that reported concentrations of these compounds are about 0.5 mM (GTP) and 0.3 mM (CTP) (15).Reports of other ATP sensors based on the epsilon subunit of ATPase do not mention the affinity of those sensors for CTP or TTP.MaLion (4) and ChemoG-ATP SiR (3) show no change in fluorescence with GTP.Finally, iATPSnFR2 is not affected by high concentrations of inorganic phosphate (SI Appendix, Fig. S1E).
MaLion has a markedly lower ΔF/F at 37 °C than at 25 °C (figure S4 of ref. 4), and ChemoG-ATP SiR is also sensitive to temperature, but less so (ext.data figure 5d of ref. 3).In contrast, iATPSnFR2 has minimal temperature sensitivity, with almost identical ΔF/F and K d for ATP at 37 °C and 25 °C (SI Appendix, Fig. S1F).The rate of fluorescence change for all three iATPSnFR2 variants is more than an order of magnitude faster than MaLionG (SI Appendix, Fig. S1G).Using the same conditions as we did for iATPSnFR2, and plotting k obs vs. [ATP], we obtain a k on for MaLionG of approximately 0.008 mM −1 s −1 , which is similar (but slightly lower) than the published value [0.011 mM −1 s −1 (figure S3 of ref. 4)].At a more practical level, this rate of binding means that MaLionG takes up to a minute to reach equilibrium when binding ATP (SI Appendix, Fig. S1H).Regardless, all three variants of iATPSnFR2 reach their maximum fluorescence within 1 to 2 s, with the A95K.S29W and A95K sensors binding ATP faster than the A95A.A119L variant.The faster kinetics of iATPSnFR2 are advantageous for observing rapid ATP fluctuations such as those occurring during neuronal firing (vide infra).
Finally, we characterized the pH dependence of iATPSnFR2.Ligand-free sensor has a pKa, near 6.0 (SI Appendix, Fig. S1H, black dashed lines).The ATP-bound form of iATPSnFR2 (SI Appendix, Fig. S1H, black solid lines) has a pKa of about 7.2.This indicates that the fluorescence of iATPSnFR2 will be quenched at very low pH in the ATP-bound state, resulting in a ~fivefold change in intensity.To avoid misinterpreting changes in fluorescence that might arise from changes in pH during experiments as changes in [ATP], we also characterized the pH profile of cpSFGFP (SI Appendix, Fig. S1H, green lines).Fortunately, its pH profile matches that of the ATP-bound state of iATPSnFR2, making it a suitable (and necessary) control for pH-induced artifacts.
Validation in Cell Lines.To validate the utility of iATPSnFR2 (and its affinity derivatives, A95K and A95K.S29W) for measuring changes in ATP production or consumption, we first tested it in two related cell lines: LMTK-, a mouse fibroblast cell line, and a Rho° (ρ 0 ) derivative, which lacks mitochondrial DNA (16).We observed changes in fluorescence in response to treatment with different ATP-affecting drugs.Because most cell lines are heavily dependent on glycolysis for energy production (17), we expected a significant decrease in fluorescence upon treatment with 2-deoxyglucose (2dOG).2dOG can be phosphorylated by hexokinase, the first enzymatic step in glycolysis, but its product (2dOG-P) cannot be further processed, resulting in competitive inhibition of phosphoglucose isomerase (18) and therefore intracellular ATP depletion.
In head-to-head experiments using a multiwell fluorescence imager, the green:red ratio of imaged cells prior to treatment was greater when the cells expressed higher affinity variants of iATPSnFR2, indicating that affinity affected ATP saturation (A95K.S29W > A95K > A95A.A119L.Fig. 1).When glucose was replaced with 2dOG, green fluorescence rapidly dropped in the ρ 0 cells (Fig. 1C), which have no alternative source of ATP other than glycolysis.The weakest affinity variant (A95A.A119L) became nearly nonfluorescent within minutes of treatment, followed by the next highest affinity variant (A95K) and finally, the highest affinity variant (A95K.S29W).In the parental LMTK-cells, 2dOG treatment resulted in decreased fluorescence only in those expressing the weakest affinity variant (A95A.A119L), indicating that alternative methods for generating ATP, namely through mitochondrial oxidative phosphorylation, were available (Fig. 1D).When the parent cells were treated with both 2dOG and oligomycin (to inhibit oxidative phosphorylation), the fluorescence of the parent cells dropped in a manner similar to the ρ 0 cells (Fig. 1E).
The self-consistent responses of the three affinity variants to 2dOG treatment illustrate their utility for cross-validation of results.In addition to the expected drop in fluorescence in response to glycolytic and oxidative phosphorylation inhibitors, Fig. 1 C-E show an increase in green fluorescence when buffers are exchanged.
To query whether this apparent increase in [ATP] was true or an artifact of using the Cytation5 multiwell fluorescence plate reader, we expressed iATPSnFR2 variants, cpSFGFP, the MaLionG ATP sensor, and eGFP in COS7 fibroblast-like cells.The process of removing buffer from the wells by pipette and then returning it results in a transient increase in fluorescence in all cells that expressed a form of circularly permuted GFP (iATPSnFR2 and cpSFGFP) or split GFP (MaLionG), while intact eGFP showed no transient increase (SI Appendix, Fig. S1I).iATPSnFR2.A95A.

Use of iATPSnFR2 to Monitor Consumption of ATP in Primary
Neurons.Neurons are highly polarized cells that consist of soma, axons, and dendrites.The generation and consumption of ATP in these different subcompartments are semiautonomously regulated.
Genetically encoded sensors can be targeted to specific subcellular compartments by using different signal sequences and can be used to monitor [ATP] during metabolic perturbations.To that end, we fused four copies of the COX8 leader sequence (19) to the N terminus of iATPSnFR2.A95A.A119L.HaloTag (mito-iATPSnFR2.HaloTag) to target it to the matrix of mitochondria.Expressing the targeted sensor in primary hippocampal neurons led to a distinct punctate appearance in both the red (visualized with JF635-HaloTag ligand) and green channels, consistent with mitochondrial targeting, and confirmed by colocalization with MitoTrackerRed (SI Appendix, Fig. S2A).Greater than 80% of mito-iATPSnFR2.HaloTag-JF635 overlaps with MitoTrackerRed, indicating that the sensor is properly localized to report mitochondrial ATP.(The fact that 20% of MitoTrackerRed does not overlap with the JF635 signal (SI Appendix, Fig. S2A) is a result of the genetically encoded sensor being sparsely transfected, while all cellular mitochondria are chemically labeled.)Additionally, we photobleached individual mitochondria and did not observe a recovery of fluorescence after photobleaching (SI Appendix, Fig. S2A), indicating that the fluorescence in mitochondria expressing mito-iATPSnFR2 arises solely from the mitochondrial compartment and is not contaminated by a cytosolic component.Application of koningic acid (KA), a covalent inhibitor of glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (20) led to a rapid drop in the ATP signal that was fully restored upon application of a mixture of lactate and pyruvate, which allows the neurons to bypass the glycolytic block by directly fueling the mitochondrial tricarboxylic acid cycle.[The JF635 signal remained unperturbed (SI Appendix, Fig. S2B)].Subsequently blocking the mitochondrial F 0 -F 1 ATPase with oligomycin led to a further collapse of the ATP signal (Fig. 2).Similar results were obtained using mito-iATPSnFR2.mIRFP670nano3(SI Appendix, Fig. S2C).We used mito-cpSFGFP.HaloTag-JF635 as a control and observed negligible changes in fluorescence upon treatment with KA or oligomycin (SI Appendix, Fig. S2D).
One key question concerning the biology of energy consumption in neurons is the nature of the mechanisms that maintain the balance between ATP consumption and ATP production during electrical activity.While this has previously been addressed by our use of luciferase (12), the time resolution of that experiment was on the order of 1 min.Here, we used iATPSnFR2.A95A.A119L.HaloTag-JFX650 to observe ATP dynamics with subsecond time resolution, during a 6-s window of intense action potential (AP) firing at nerve terminals of dissociated hippocampal neurons in culture (Fig. 3).When neurons were stimulated in buffer containing 5 mM glucose, the fluorescence dropped slightly, and recovered to baseline within 1 min of concluding the stimulation.When we subsequently inhibited production of ATP by glycolysis with addition of KA, we observed an even greater decrease in fluorescence that did not recover within the minute and a half observation window.The higher affinity variant iATPSnFR2.A95K.HaloTag-JFX650 was still able to detect stimulusdependent changes in fluorescence, albeit with a much smaller amplitude.The negative control cpSFGFP (without an ATP-binding domain) remained unchanged even after ATP depletion.iATPSnFR1, which has a lower ΔF/F than iATPSnFR2, did not significantly change fluorescence during the 300 AP stimulation protocol, with or without KA treatment, although the endpoint of treatment with KA did show a slight drop in fluorescence (SI Appendix, Fig. S3A).We found that MaLionG, which has slower kinetics than iATPSnFR2 (see above), was not well suited for detecting changes on these time scales.Following acute electrical stimulation in a buffer containing KA, the MaLionG signal reported a statistically insignificant change in fluorescence of only 9% (SI Appendix, Fig. S3C), compared to a ~37% drop using iATPSnFR2.The data obtained using iATPSnFR2 recapitulate our previous observations that nerve terminals have a robust on-demand ATP synthesis program that responds rapidly to electrical activity (10), while improving the time resolution by an order of magnitude.
iATPSnFR2 Reveals Differences in ATP Consumption among Individual Boutons.We and others have previously shown that, even in the absence of electrical activity, nerve terminals have high basal ATP consumption.However, the temporal and spatial resolution was insufficient for resolving kinetic details at the single-bouton level, due to either the low photon flux of the luciferase-based reporter (21) or the limited dynamic range of the FRET sensor-based reporter ATeam (22).We repeated these experiments using synaptically targeted iATPSnFR2 that was fused to synaptophysin (23), sending it to the synaptic vesicle surface, facing the cytosol.Similar to previous findings, when ATP production was blocked (in this case by inhibiting GAPDH with KA) in the presence of the Na + channel blocker  tetrodotoxin (TTX), the average to red fluorescence ratio of iATPSnFR2.A95A.A119L.HaloTag-JFX650 (averaged over 7 different neurons, each contributing 20 to 30 boutons) gradually declined over a 5 to 10 min period as previously reported (21).The weaker A95A.A119L variant showed a greater drop than the medium affinity A95K variant (Fig. 4A).In contrast to what we observed in the mitochondrially targeted sensor, when using JF635 as a HaloTag ligand, the red channel fluorescence was not inert with respect to ATP changes (SI Appendix, Fig. S4A).Rhodamine-based dyes can interact with ATP on their own (24), and it appears that HaloTag-JF635 itself is sensitive to ATP, albeit with low ΔF/F (SI Appendix, Fig. S4B).This confounding artifact was minimized by using a newer-generation HaloTag dye, JFX650 (SI Appendix, Fig. S4A).The improved signal to noise properties of iATPSnFR2 allowed us to extract several features regarding single-synapse ATP control.Comparisons of the initial ratio values across synapses and experiments showed that across boutons, the resting ATP value varied as much as fourfold, and by 50% across cells, but this variation was minimized following ATP depletion by KA (Fig. 4B).Similarly, the kinetics of ATP depletion in individual nerve terminals behaved unexpectedly.Although the average behavior across all boutons shows a gradual decline in ATP (Fig. 4A), no individual bouton looked like the average, displaying instead two distinct kinetic features that varied widely across boutons (Fig. 4C).At the single-bouton level, ATP signals declined very rapidly (typically showing depletion within ~10 to 20 s) but only after a delay time that varied significantly across boutons.Individual bouton ATP levels under these conditions were best described by a simple Boltzmann function, F(t) = F0/ (1 + e 4 α (t−t ½ ) ) where t ½ describes the delay time to the precipitous drop in signal at a rate α (Materials and Methods) (Fig. 4D).Both these kinetic parameters, t ½ and α, varied significantly across the population of nerve terminals.A correlation analysis of the two extracted parameters showed that, in general, smaller t ½ were associated with higher values of α (Fig. 4E) and vice versa.Some cells showed tighter clustering of rate and t ½ parameters than others (Fig. 4F).There was no correlation between decay rate or t ½ and expression of the sensor (SI Appendix, Fig. S4C).Neither the initial nor final green fluorescence values (before and after the KAinduced crash) of individual boutons appear to correspond to the expression level of the sensor, as determined by JFX650 brightness (SI Appendix, Fig. S4D), indicating that the observed diversity of ATP concentrations and consumption rates in individual boutons is genuine, and not an artifact of sensor expression levels.

Discussion
Here, we present an improved ATP sensor-iATPSnFR2, optimized for both higher fluorescence changes in response to saturating ATP (ΔF/F ~ 12), and modulation of its affinity for ATP such that its range of maximal sensitivity (K d ) matches the expected range of ATP concentrations within normally functioning cells, including neurons.We also provide two higher affinity variants to serve as controls or for use in other cell types or cellular subcompartments that might be expected to have lower concentrations of ATP.Since iATPSnFR2, like other cpGFP-based sensors, has greater pH dependence than intact, FRET-based sensors, such as the original ATeam ( 6), we also use a cpSFGFP without the ATP-binding component to serve as a control for induced changes in pH.For other experimenters interested in using iATPSnFR2 to investigate changes in energy metabolism, we strongly recommend using the cpSFGFP control as well, especially when using the mitochondrially targeted sensors, since depolarization of the mitochondrial membrane can reduce pH by an entire pH unit (25).We validate its utility for measuring changes in cytosolic ATP using cellular imaging in cultured cell lines, perturbed with different pharmacological compounds known to affect ATP production and consumption.Inclusion of a far-red fluorophore (via HaloTag or mIRFP670nano3) allows for the option of ratiometric measurement to account for expression and movement artifacts.iATPSnFR2 can be targeted to the axonal mitochondria, where it reports homeostasis of mitochondrial ATP in the presence of a suitable fuel for the tricarboxylic acid cycle.It also reports depletion and subsequent rescue of mitochondrial ATP after perfusion of oxidative phosphorylation substrates and inhibitors.Finally, by targeting it to boutons, we observed spontaneous depletion of ATP with each bouton having its own unique ATP consumption profile.
While multiple other fluorescence-based sensors have been developed, none of them have been widely adopted for determination of ATP consumption in cell culture or in vivo.The ATP production pathway through glycolysis produces lactate and acidifies the extracellular space.ATP production from the electron transport chain in mitochondria consumes O 2 .The extracellular acidification rate and oxygen consumption rate can be measured and used as indirect proxies for ATP consumption and are the underlying technologies of the Seahorse analyzer (26).Typically, these rates are measured in cell culture under normal/healthy conditions to establish a baseline, and then, mitochondrial inhibitors (oligomycin and rotenone/antimycin A, but sometimes other inhibitors) are added to drop mitochondrial ATP production to zero, and through a series of complex calculations, ATP production rates, divided between mitochondrial and glycolytic production, are determined.
Measuring pH and O 2 consumption in cell culture is useful for determining the energetic expenditure differences between cell types or between cell cultures that have been treated with different drugs, but the technique also has significant limitations.Primarily, measuring pH changes and O 2 consumption are just proxies for ATP production, and not a direct measurement.Other metabolic events or perturbations that affect pH and O 2 use will also be interpreted as ATP consumption.Importantly, the Seahorse system limited to cell culture analysis and is not compatible with in vivo applications or measurement of ATP consumption with other manipulations, such as electrical stimulation.This technology also affords no spatial resolution either at the single cell or subcellular length scale.Furthermore, once cells have been treated with mitochondrial poisons, they are no longer viable and cannot be assayed again.
Finally, the time course of measurement using such devices requires integration of signals over time and is not compatible with subsecond resolution.We expect that iATPSnFR2 will provide researchers with opportunities to study ATP dynamics with temporal and spatial resolution that has, until now, been unobtainable.

Materials and Methods
Mutagenesis and Screening of Bacterial Lysates.iATPSnFR1 was cloned into a bacterial expression vector derived from pRSET (Invitrogen) with restriction sites designed to make downstream subcloning into mammalian expression vectors easier.Site saturation mutagenesis was performed using the uracil template method (27), as described previously (28).Mutagenesis reactions were transformed into T7 express bacterial cells (New England Biolabs) and plated on LB-Amp agar plates.Individual colonies were picked into 2 mL, square well, 96well plates filled with 0.9 mL autoinduction media (29) containing ampicillin and grown overnight at 30 °C with rapid shaking (350+ RPM).To remove excess ATP, 96-well plates were centrifuged, pellets resuspended in Tris Buffered Saline, and recentrifuged, a total of 3 times, then frozen overnight as dry pellets.iATPSnFR variants were assayed in bacterial lysate by addition of 0.9 mL Mammalian Cell Imaging Buffer (20 mM HEPES, 119 mM NaCl, 2.5 mM KCl, 2 mM MgCl 2 , and 2 mM CaCl 2 , pH 7.2), rapid vortexing, and then pelleting by centrifugation.Clarified lysate was transferred to a black 96-well plate (Greiner).Fluorescence was measured in a plate reader (Tecan Spark) with 485 nm excitation (20 nm bandpass) and 535 nm emission (20 nm bandpass).ATP (Sigma) was added to 1 µM and fluorescence remeasured.ATP was added to 1 mM and fluorescence remeasured.Variants with responses to either concentration with high ΔF/F were carried forward with mutations of linker 1 until one of those winners, L1-VLVG was sufficient to become the template for mutations at linker 2.
Protein Expression and Purification and Characterization.iATPSnFR2 variants were transformed into Escherichia coli BL21(DE3) pLysS cells.Protein expression was induced by growth in 300 mL autoinduction media (29) supplemented with 100 µg/mL ampicillin at 30 °C.Proteins were purified by immobilized Ni-NTA affinity chromatography (30).iATPSnFR2 proteins were eluted with a 120-mL gradient from 0 to 200 mM imidazole.Fractions that were fluorescent were pooled and concentrated by ultrafiltration (Amicon) and then dialyzed by Slide-A-Lyzer cassette (30 kDa cutoff) in Mammalian Cell Imaging Buffer.Protein concentration was quantified by alkali denaturation and measurement of the GFP chromophore, with an extinction coefficient of 44,000 M −1 cm −1 at 447 nm.
Fluorophore Conjugation.Purified protein was mixed with 1.1 molar equivalents of JFX650 HaloTag ligand and incubated at room temperature for 1 h and then at 4 °C overnight.Excess HaloTag ligand was removed by gel filtration on a PD-10 column (GE Life Sciences) and reconcentrated by ultrafiltration.
Equilibrium Measurements.All in vitro assays were performed in Mammalian Cell Imaging Buffer unless otherwise noted.iATPSnFR2 equilibrium measurements (affinity, specificity, pH) were performed with 0.2 µM protein in a Tecan Spark fluorimeter with 20 nm bandpass windows and excitation at 485 nm and emission at 535 nm.Kinetic Measurements.iATPSnFR2 protein (0.2 µM) was rapidly mixed with an equal volume of stock ATP solution in an Applied Photophysics SX-20 stopped flow fluorimeter with a 490 nm LED excitation and 510 nm long pass filter for emission.The final concentration of protein was thus 0.1 µM.Concentrations of ATP listed in figures are the final concentrations after mixing.
Cloning into Mammalian Expression Vectors.iATPSnFR2.HaloTag variants were subcloned by restriction digest into an AAV vector with a CAG promoter via BglII/ PstI digest from the bacterial expression vector.The pAAV.CAG vector includes an extra serine after the initial methionine and lacks a polyhistidine tag.To target the sensor to the mitochondrial matrix, four repeats of COX8(mito) leader sequence (SVLTPLLLRGLTGSARRLPVPRAK) separated by spacer (IHSLPPEGPW) were introduced at the N terminus of iAPTSnFR2(A95A/A119L).HaloTag was incorporated at the N-terminal of iATPSnFR2.A95A.A119L separated by a linker (LQSTGSGNAVGQDTQER).The plasmid was transformed into stbl3 competent E. coli, and DNA was harvested from 200 mL growth media using an endotoxin-free plasmid purification kit (Qiagen).
Rho° Cell Culture.Rho° and the parental LMTK-cells were maintained in T-75 flasks with Dulbecco's Modified Eagle's Medium (DMEM) + 10% Fetal Bovine Serum (FBS) + 1 mM pyruvate + 5 µM uridine.Two days prior to transfection, they were split into separate rows of a 24-well plate (half Rho°, half LMTK) at 0.6 and 0.1 million cells per well, respectively.Cells were transfected with iATPSnFR2.HaloTag variants by lipofectamine (0.5 µg DNA + 2 µL lipofectamine 2,000 per well) for 4 to 6 h in DMEM lacking FBS or antibiotics.Transfection medium was replaced with fresh DMEM + 10% FBS + penicillin/streptomycin overnight.Cells were labeled with JFX650 HaloTag ligand by adding the fluorophore to 100 nM for an hour.Cells were washed with Mammalian Cell Imaging Buffer and imaged in a Cytation 5 imaging reader (BioTek) over the course of about 2 h.After about 20 min of imaging to establish a baseline fluorescence, the culture plate was removed from the instrument, and buffer was replaced with the query buffer.The equilibration buffer contained 10 mM glucose, and the query buffers contained 10 mM 2dOG.
Rho° Cell Image Processing.Images were background subtracted and aligned for movement artifacts with the Gen5 software package that controls the Cytation 5 instrument.Images were assembled as TIF stacks and imported into ImageJ.For each well, an automated, custom script identified regions of interest (ROIs) by thresholding on the far-red channel and expanding by 2 pixels.Those ROIs were used to determine the ratio of green to red fluorescence for each well.ROIs were curated to remove those that outlined cells that detached or appeared to become spherical during the experiment.

Animals.
All experiments involving animals were performed in accordance with protocols approved by the Weill Cornell Medicine Institutional Animal Care and Use Committee.Neurons were derived from Sprague-Dawley rats (Charles River Laboratories strain code: 001, RRID: RGD_734476) of either sex on postnatal days 0 to 2).
Synaptic Bouton and Axonal Mitochondria Image Analysis.Time series of imaging pairs (HaloTag and iATPSnFR2) were split into two independent image series using a custom-written Fiji routine to facilitate analysis.ATP signals are reported as a ratio between iATPSnFR2:HaloTag (Green:Red).Images were analyzed using the ImageJ plug-in Time Series Analyzer V3 where 50 to 100 circular ROIs of radius 1 µm corresponding to synaptic boutons or mitochondria expressing the Syn-iATPSnFR2.HaloTag or mito-iATPSnFR2.Halo (Halo dye positive) were selected.Image loading and posterior raw data saving were automatized using a homemade Python code for Fiji.Synaptic boutons signals were analyzed using homemade script routines in Igor-pro v6.3.7.2 (WaveMetrics, Lake Oswego, OR, USA).ATP ratio signal (Green:Red) was calculated per individual bouton, normalized to the baseline, and fitted with a Boltzmann's sigmoidal function from the time that 10 µM KA was administered, as follows: For electrical activity experiments, single-cell data were filtered by excluding any cell data where no ATP change was observed when cells were stimulated in the presence of KA.
Data, Materials, and Software Availability.Raw data used to generate figures have been deposited in FigShare (DOI: 10.6084/m9.figshare.25647993)(33).All study data are included in the article and/or SI Appendix.
A119L shows the largest increase in fluorescence with a multiminute return to baseline, indicating that the physical perturbation of buffer exchange might induce a cytosolic increase in ATP production (or decrease in ATP consumption).(Note: A transient apparent change in [ATP] upon treatment is also observed with the ChemoG-ATP SiR sensor (figure 3e of ref. 3).

Fig. 3 .
Fig. 3. Depletion of cytosolic ATP in nerve terminals during AP firing.(Bottom Left) Ratio of green to red fluorescence of axonal terminals in cultured neurons expressing iATPSnFR2.HaloTag-JFX650 during a burst of AP firing [6 s at 50 Hz (gray bar)] in either 5 mM glucose (dashed lines) or 5 mM glucose + 10 µM KA (solid lines).Low affinity variant A95A.A119L in yellow (n = 12) and medium affinity variant A95K in orange (n = 7).Identical experiments carried out using cpSFGFP control (green) showed no change for the same stimulus for either condition (n = 6).(Top) Images at the completion of the experiment in both green and red channels; (scale bar: 10 µm.) (Bottom Right) Maximum change in ratio during the stimulation period in the absence of KA (open circles) or presence of 10 µM KA (filled circles).***P < 0.001, *P < 0.05, n.s.P > 0.05 paired t test.

Fig. 4 .
Fig. 4. Spontaneous depletion of ATP in boutons detected by iATPSnFR2.iATPSnFR2 was targeted to boutons by fusing it to the C-terminal of synaptophysin.(A) Average ratio of green to red fluorescence of iATPSnFR2-HaloTag-JFX650 normalized to pretreatment baseline.Yellow, low affinity variant A95A.A119L (n = 7); orange, medium affinity variant A95K (n = 6), ± SE.Treatment with 5 mM glucose + 10 µM KA + 0.3 µM TTX initiated at t = 0. (B) Variation in ATP across individual boutons (n = 577) and across individual neurons pre-and post-ATP depletion by KA of the 7 cells shown in A with the low affinity variant.(C) Representative raw traces from individual boutons expressing A95A.A119L variant.(D) Normalization of representative traces illustrates that individual boutons have unique half-times (t ½ ) to ATP depletion and unique rates of ATP consumption.(E) Cross-correlation of single-bouton depletion rates vs. t ½ values shows that there is in general an inverse relationship between these parameters, with some cells occupying regions of the parameter space.(F) Boutons from only two cells (cells 4 and 7 in B) shown for clarity.
1/2 is the t value where Y is at (base + max)/2, and α represents the rate of signal drop derived from , where t